Journal of Educational Psychology Copyright 1999 by the American Psychological Association, Inc.

1999, Vol. 91, No. 3,497-510 0022-0663/99/S3.00

The Domain Specificity and Generality of Belief Bias:

Searching for a Generalizable Critical Thinking Skill
Walter C. Sa Richard F. West
University of Toronto James Madison University

Keith E. Stanovich
University of Toronto

The domain specificity and generality of belief-biased reasoning was examined across a height
judgment task and a syllogistic reasoning task that differed greatly in cognitive requirements.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Moderate correlations between belief-bias indices on these 2 tasks falsified an extreme form of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

the domain specificity view of critical thinking skills. Two measures of cognitive ability and 2
measures of cognitive decontextualization skill were positively correlated with belief bias in a
height judgment task where prior knowledge accurately reflected an aspect of the environment
and negatively correlated with belief bias in a height judgment task where prior knowledge
was incongruent with the environment. Likewise, cognitive ability was associated with skill at
resisting the influence of prior knowledge in the syllogistic reasoning task. Participants high in
cognitive ability were able to flexibly use prior knowledge, depending upon its efficacy in a
particular environment. They were more likely to project a relationship when it reflected a
useful cue, but they were also less likely to project a prior belief when the belief was
inefficacious.

An important research tradition within cognitive science
has demonstrated how prior belief biases the evaluation of
arguments and of data (Baron, 1995; Evans, Over, &
Manktelow, 1993; George, 1995; Klaczynski, Gordon, &
Fauth, 1997; Klaczynski & Narasimham, 1998; Moshman &
Franks, 1986). The quintessential paradigm for demonstrat-
ing this phenomenon is the syllogistic reasoning problem,
which pits the believability of the conclusion against the
validity of the argument (e.g., Evans, Barston, & Pollard,
1983; Markovits & Nantel, 1989). In this paradigm, the
belief bias effect occurs when participants are found to judge
the validity of the syllogism more accurately when the
believability of the conclusion coincides with the validity of
Walter C. S£ and Keith E. Stanovich, Department of Human
Development and Applied Psychology, University of Toronto,
Ontario, Canada; Richard F. West, Department of Psychology,
James Madison University.
Walter C. Sa is now at the Department of Psychology, Grand
Valley State University.
This research was supported by a grant from the Social Sciences
and Humanities Research Council of Canada. Preparation of the
manuscript was supported by a Connaught Research Fellowship in
the Social Sciences, awarded to Keith E. Stanovich.
We thank Monica Biemat for generously providing the photo-
graphs from her 1990 study. We also thank Michael Doherty and
Denis Hilton for their comments on an earlier version of the
manuscript.
Correspondence concerning this article should be addressed to
Keith E. Stanovich, Department of Human Development and
Applied Psychology, Ontario Institute for Studies in Education,
University of Toronto, 252 Bloor Street West, Toronto, Ontario,
Canada M5S 1V6. Electronic mail may be sent to kstanovich
@ oise.utoronto.ca.
the syllogism than when it conflicts. For example, problems
that are invalid and have unbelievable conclusions (e.g., All
guns are dangerous. Rattlesnakes are dangerous. Therefore,
rattlesnakes are guns.) are easier than problems that are
logically invalid and have believable conclusions (e.g., All
living things need water. Roses need water. Roses are living
things.). Presumably, in the latter situation, the evaluation of
logical validity is disrupted by the real-world knowledge
that is cuing an alternative response. Prior belief (in this
case, knowledge of the world) thus interferes with optimal
task performance, which is attained by attending only to
logical validity.
Belief bias based on prior knowledge of real-world
relationships has also been demonstrated in other paradigms
such as the evaluation of numerical covariation information
(Broniarczyk & Alba, 1994). Additionally, belief bias, based
not on prior knowledge but on opinion about a controversial
issue, has been demonstrated in several paradigms such as
covariation detection (Levin, Wasserman, & Kao, 1993;
Stanovich & West, 1998b), argument evaluation (Klaczyn-
ski, 1997; Stanovich & West, 1997, 1998a), and evidence
evaluation (Kardash & Scholes, 1996; Klaczynski & Gor-
don, 1996; Klaczynski et al., 1997; Klaczynski & Narasim-
ham, 1998; Kuhn, 1991; Slusher & Anderson, 1996).
Similarly, the social psychological literature contains many
demonstrations of how the evaluation of communications is
biased by prior opinion and belief (e.g., Biek, Wood, &
Chaiken, 1996; Dole & Sinatra, 1998; Kunda, 1990; Lord,
Lepper, & Preston, 1984; Nickerson, 1998; Wegener &
Petty, 1997). Belief biases of these various types exemplify
what Stanovich (1999) has termed the fundamental computa-
tion bias of human cognition, which refers to a bias that arises
because of difficulties in cognitive decontextualization.

497
 498 SA, WEST, AND STANOVICH
As in the syllogistic reasoning literature, all of these
effects represent cases where prior knowledge, belief, and
opinion serve to disrupt the impartial evaluation of evidence
and argument. The ability to avoid this type of belief bias is
repeatedly stressed as a positive trait in the literature on
critical thinking. For example, Norris and Ennis (1989)
argued that one important characteristic of critical thinking
is the tendency to "reason from starting points with which
they disagree without letting the disagreement interfere with
reasoning" (p. 12). Zechmeister and Johnson (1992) listed as
one characteristic of the critical thinker the ability to "accept
statements as true even when they don't agree with one's
own position" (p. 6). Similarly, Nickerson (1987) stressed
that critical thinking entails the ability to recognize "the
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
fallibility of one's own opinions, the probability of bias in
those opinions, and the danger of differentially weighting
evidence according to personal preferences" (p. 30). This
sentiment has been echoed by many critical thinking theo-
rists (e.g., Baron, 1991, 1995; Baron & Sternberg, 1987;
Klaczynski & Narasimham, 1998; Kuhn, 1991,1996; Mosh-
man, 1994; Perkins, 1995; Perkins, Jay, & Tishman, 1993;
Siegel, 1988, 1997).
In these statements championing the avoidance of belief-
biased reasoning, the ability to evaluate evidence in an
unbiased manner has been treated as a global trait. For
example, when Nickerson (1987) warned against "differen-
tially weighting evidence according to personal prefer-
ences" (p. 30), the domain of the preferences is not
specified. Likewise, when Zechmeister and Johnson (1992)
championed the ability to "accept statements as true even
when they don't agree with one's own position" (p. 6), they
did not specify a statement domain. Throughout the entire
critical thinking literature, domain generality has been
assumed for the various thinking styles that are listed as the
defining features of critical thinking (e.g., Ennis, 1987;
Lipman, 1991; Wade & Tavris, 1993). Indeed, Baron
(1985b) made some degree of domain generality a defining
feature of his notion of a thinking style: "Cognitive styles
ought to be general. By ought I mean that evidence against
the generality of a style is taken to make the style less
interesting" (pp. 379-380). This view leads to the obvious
individual difference prediction that "we should expect
some correlation across individuals between style in one
situation and style in another, regardless of how discrepant
the situations are" (Baron, 1985b, p. 380). However, there
have been virtually no such multivariate studies reported on
the belief bias effect. Whether the phenomenon has any
degree of domain generality or whether it is domain specific
(and thus, under Baron's criterion, does not warrant its
treatment as a thinking style in the critical thinking litera-
ture) is almost completely unknown.
In contrast to the assumption of domain generality in the
critical thinking literature, in developmental and cognitive
psychology, the reigning assumption for at least a decade has
been one of domain specificity. For example, the contextual-
ist tradition within developmental psychology emphasizes
the point that the exercise of cognitive skills is often quite
situation-specific (J. S. Brown, Collins, & Duguid, 1989;
Lave & Wenger, 1991; Rogoff & Lave, 1984). Likewise,
theorists who have emphasized the importance of domain
knowledge (e.g., Alexander, Kulikowich, & Schulze, 1994;
Ceci, 1993, 1996) have argued that many so-called basic
cognitive processes are so dependent on familiarity with the
specific stimulus domain and its context that it seems almost
a misnomer to call them "basic" (see Ceci, 1996). Stigler
and Baranes (1988) stressed this point in particular by
arguing that "it is more and more difficult, within modern
cognitive theory, to draw a division between basic processes
and 'mere' content" (p. 257).
Thus, theorists in both the domain-knowledge tradition
and the contextualist tradition of developmental and educa-
tional psychology emphasize the domain specificity of the
exercise of a cognitive skill or style (for a critique of this
view, see J. R. Anderson, Reder, & Simon, 1996). Such
theorists question the existence of thinking styles that have
the generality ascribed by Baron (1985b). For example,
under a contextualist conceptualization, belief bias effects
would be expected to display extreme domain specificity.
General tendencies to avoid belief bias, such as those
discussed in the critical thinking literature, are thus treated
with extreme skepticism within the contextualist framework.
That there are influential but competing traditions within
which belief bias effects can be conceptualized (the critical
thinking literature, the contextualist tradition, and the domain-
knowledge tradition) highlights the urgency of producing
empirical data that can at least partially adjudicate the
differential predictions. In the present article, we present one
of the few attempts to examine the domain generality of
belief bias as a cognitive style cutting across tasks that
widely differ in cognitive and response requirements.
Research Strategy
In the present study, we compared two tasks on which
belief bias has been displayed. One task, typical of verbal
reasoning paradigms, is the syllogistic reasoning task dis-
cussed previously. The second task was drawn from the
perceptual judgment domain in order to contrast with the
verbal reasoning domain of the syllogistic reasoning task.
This task was adapted from the work of Nelson, Biernat, and
Manis (1990) who had participants judge the heights of
seated and standing males and females in photographs. They
found that the judgments were related to the actual heights,
but that the over- and underestimations of the actual heights
were related to the gender of the person in the target
photograph. That is, although gender is a valid cue in that it
is related to actual height, participants tended to overproject
this cue. Importantly, this over projection was maintained in
a condition (termed the matched condition) where the
participants were (a) informed that the male and female
pictures were matched in height, and (b) warned not to use
gender as a cue.
From the standpoint of the present investigation, the
matched condition of the Nelson et al. (1990) height
judgment task is important because it mirrors the logic of the
syllogistic reasoning task. In both cases, the participant is
 BELIEF BIAS
told to ignore a cue that is normally diagnostic but in this
particular situation is not predictive (the truth status of the
conclusion in the syllogistic reasoning task and the gender of
the stimulus in the height judgment task). In both cases, it
has been found that, in the aggregate, people are unable to
ignore the nondiagnostic cue and evaluate only the informa-
tion of interest (the validity of the syllogism and the valid
height cues in the photograph, respectively). These tasks are
particularly interesting for investigating issues of domain
specificity and generality because, despite the similarity of
the belief bias logic across the two, the stimuli used in the
tasks are very different, and the judgments are also quite
different (one is a judgment of logical validity and the other
a height judgment).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Thus, in the present investigation, in addition to a
This document is copyrighted by the American Psychological Association or one of its allied publishers.
syllogistic reasoning task, we examined two versions of the
Nelson et al. (1990) height judgment paradigm. We com-
pared a situation where ecologically valid stimuli were used
with one where the stimuli that were chosen conflicted with
prior belief. In the former—termed the ecological condition—
the photographs viewed by the participants were a reason-
ably representative sample of the actual heights of males and
females in North America, and participants were informed
of this fact. In the latter condition—the matched condition—
participants viewed a sample of photographs where the
mean heights of the males and females were matched, and
participants were informed of the matching (that is, they
were informed that the gender cue was nondiagnostic). This
condition is the one that is structurally analogous to the
syllogistic reasoning conflict condition.
In our study questionnaire, we included measures of
styles of epistemic regulation that embody the fundamental
assumption of the critical thinking literature—that actively
open-minded thinking is exercised in a domain-general
manner (Baron, 1993, 1994; Perkins et al., 1993). We
constructed our questionnaire by drawing upon scales al-
ready in the literature and by constructing some of our own.
Several of our measures had strong similarities to other
related measures in the literature (see Klaczynski et al.,
1997; Schommer, 1990, 1993, 1994; Schommer & Walker,
1995; Webster & Kruglanski, 1994). Perhaps the strongest
similarities are with the two factors that Schommer (1990,
1993) called belief in simple knowledge and belief in certain
knowledge. That is, we particularly focused on thinking
styles that had potential epistemic significance (see Ceder-
blom, 1989); for example, "the disposition to weigh new
evidence against a favored belief heavily (or lightly), the
disposition to spend a great deal of time (or very little) on a
problem before giving up, or the disposition to weigh
heavily the opinions of others in forming one's own"
(Baron, 1985a, p. 15). We attempted to tap the following
dimensions: epistemological absolutism, willingness to
switch perspective, willingness to decontextualize, identifi-
cation with beliefs, and the tendency to consider alternative
opinions and evidence. The focus with respect to these
measures was on whether they predicted the actual magni-
tude of the belief bias effect and whether that predictive
validity was domain general.
We included converging measures of cognitive ability in
499
order to examine the type of processing to which the human
cognitive apparatus is more prone when freed of computa-
tional limitations. If the received position in the critical
thinking literature is correct, people with greater cognitive
and reasoning ability should show a greater tendency to
evaluate evidence and arguments independent of prior belief
(i.e., they should display the cognitive flexibility necessary
to detach reasoning from prior knowledge). In contrast, if
contextualist theories are correct, tendencies toward decon-
textualization might be viewed as inefficacious, and people
with increased cognitive ability might display greater reluc-
tance to detach world knowledge from their information
processing (i.e., they will demonstrate increased levels of
belief bias). Finally, we included two other tasks that require
cognitive decontextualization in order to test the domain
generality of styles of epistemic regulation.
Method
Participants
The participants, who consisted of 124 students (54 males and 70
females), were recruited at a large Canadian university through
poster advertisements that were distributed on campus. Their mean
age was 21.8 years (SD = 3.7), and the modal age was 19.
Participants were paid $20 per student for their cooperation.
Belief Bias Tasks
Height judgment task. The height judgment task was adapted
from the paradigm used by Nelson et al. (1990). Participants were
presented with full-length photographs of males and females, who
were seated in a variety of natural settings. The participants' task
was to judge as accurately as possible the heights of the individuals
in each of these photographs. The stimuli in the first height
judgment task—termed the ecological height judgment task—were
sampled to reflect the actual male and female heights in the
population. These stimuli consisted of 83 pictured models, 40
males and 43 females. Seventy-five of these came from the Nelson
et al. (1990) study and were generously provided by those
investigators. The mean height of the males in the photographs was
5 ft, 10.6 in. (SD = 2.7 in.), and the mean height of the females in
the photographs was 5 ft, 5.6 in. (SD = 3.4 in.). These mean
heights are roughly equal to those of the general population to
which the participants belonged (National Dairy Council, 1980).
The photographs were taken at varying distances. Thus, there was
no significant correlation between image size and actual height
(r=-.159).
The photographs were presented in a randomized order, mixing
males and females. The instructions in the ecological height
judgment task were as follows:
This task is designed to assess how accurately people can
judge the physical characteristics of individuals based on a
small amount of information. You will be looking at photo-
graphs of individuals and will be asked to judge the height of
each person pictured. The task is fairly difficult because the
models are all seated. On each page of this booklet you will
find a picture of a person. Your task is simply to look at each
photograph and estimate the height of the pictured person.
Each person's height was measured when they were wearing
the shoes that they are wearing in the picture. Therefore, your
height estimate should be an estimate of the person's height
 500 SA, WEST, AND STANOVICH
while wearing the shoes in the picture. As you flip through
these pictures, read out loud to the experimenter your height
estimate.
Participants were allowed to give their estimates either in
centimeters or in feet and inches (the vast majority preferred the
latter). The judgments of the few participants who gave centimeter
estimates were transformed into feet and inches.
The second version of the height judgment task—termed the
matched height judgment task—immediately followed the ecologi-
cal version. In this version, participants were specifically informed
that the gender cue had been rendered nondiagnostic and conse-
quently should be ignored in their ensuing height judgments.
Participants estimated the heights of 46 pictured models (23 male
and 23 female), where the mean male and female heights were
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
equated at 5 ft, 8.5 in. (SDs = 3.4 and 3.6, respectively). Again, the
This document is copyrighted by the American Psychological Association or one of its allied publishers.
stimuli came from the Nelson et al. (1990) study and were
presented in a randomized order, mixing males and females. All of
these 46 stimuli were different from those used in the ecological
condition. The instructions in the matched height judgment task
were as follows:
In the following task, you will be doing exactly what you were
just doing before. You will again be looking at several pictured
people and estimating their height. This time, however, the
men and women pictured are, on average, of equal height.
That is, we have taken care to match the heights of the men
and women pictured (as before, footwear is included in the
measurement). For every woman of a particular height,
somewhere in this section there is also a man of that same
height. Therefore, in order to make as accurate a height
judgment as possible, try to judge each photograph as an
individual case. Do not rely on the person's sex to aid your
judgment.
One participant failed to complete these tasks and 6 were
removed as multivariate outliers (having Mahalanobis distance
metrics greater than 4.0; see Tabachnick & Fidell, 1983).
Syllogistic reasoning task Twenty-four syllogistic reasoning
problems, largely drawn from Markovits and Nantel (1989), were
completed by the participants. Eight of the problems were worded
such that the validity judgment was in conflict with the believabil-
ity of the conclusion (e.g., All living things need water; Roses need
water; therefore, Roses are living things—which is invalid). These
were termed the inconsistent items. Eight of the problems were
worded such that the validity judgment was congruent with the
believability of the conclusion (e.g., All fish can swim; Tuna are
fish; therefore, Tuna can swim—which is valid). These were
termed the consistent items. Eight of the problems involved
imaginary content (e.g., All opprobines run on electricity; Jamtops
run on electricity; therefore, Jamtops are opprobines—which is
invalid). These were termed the neutral items. The instructions that
were given to the participants were as follows:
In the following problems, you will be given two premises
which you must assume are true. A conclusion from the
premises then follows. You must decide whether the conclu-
sion follows logically from the premises or not. You must
suppose that the premises are all true and limit yourself only
to the information contained in the premises. This is very
important. Decide if the conclusion follows logically from the
premises, assuming the premises are true, and circle your
response.
After each item, the participants indicated their responses by
circling one of two alternatives: (a) "Follows Logically," or (b)
"Does Not Follow Logically."
Cognitive Decontextualization Tasks
Covariation judgment. For this task, we adapted a paradigm
where people were presented with covariation information that was
accommodated by the format of a 2 X 2 contingency table (see
Schustack & Sternberg, 1981; Wasserman, Dorner, & Kao, 1990).
Participants indicated their opinions on 25 sets of hypothetical
relationships covering a variety of different topics (e.g., that
couples who live together before marriage tend to have successful
marriages) and then evaluated the degree of association between
the same 25 sets of two variables in the data of a hypothetical
research study. These data corresponded to the four cells of the 2 X
2 contingency table, traditionally labeled A, B, C, and D (see Levin
et al., 1993). For example, in the "living together" problem,
participants were told that a researcher had sampled 225 couples
and had found that (a) 100 couples lived together and had
successful marriages, (b) 50 couples lived together and were
divorced, (c) 25 couples did not live together and had successful
marriages, and (d) 50 couples did not live together and were
divorced. Subsequent to the presentation of one set of these data,
the participants were asked to judge the nature and extent of the
relationship between living together before marriage and success-
ful marriages in these data on a scale ranging from +10 (positive
association) to —10 (negative association) and centered on 0 (no
association). Details of the values used in the tables were presented
in Stanovich and West (1998b).
The normatively appropriate strategy in this task (see Allan,
1980; Kao & Wasserman, 1993; Shanks, 1995) is to use the
conditional probability rule (i.e., subtracting the probability of the
target hypothesis when the indicator is absent from the probability
of the target hypothesis when the indicator is present. Numerically,
the rule amounts to calculating the Ap statistic: [A/(A + B)] —
[C/(C + D)], (see Allan, 1980). Thus, each participant's 25 covaria-
tion scores were regressed simultaneously on the 25 Ap values and
the 25 prior-opinion scores. The former beta weight was used as the
primary indicator of the ability to reason about covariation
independently of their prior opinion on the issue in question.
Argument evaluation test. The second reasoning task was
analogous in structure to the covariation judgment task but
involved the assessment of argument strength rather than numerical
covariation. The Argument Evaluation Test (AET; see Stanovich &
West, 1997, for details) consisted of two parts. First, participants
indicated their degree of agreement with a series of 23 target
propositions (on topics such as gun control, taxes, university
governance, crime, etc.) on a 4-point scale. Participants then
evaluated arguments (which varied on an operationally defined
measure of strength; see Stanovich & West, 1997) relevant to these
propositions. Individual differences in participants' reliance on
objective argument quality were examined by running separate
regression analyses on each participant's responses. Each partici-
pant's 23 argument-evaluation responses were regressed simulta-
neously on both the 23 argument-quality scores and the 23
prior-opinion scores. The former beta weight was used as the
primary indicator of the participants' ability to evaluate arguments
independently of their prior opinion on the issue in question.
Cognitive Ability Measures
Cognitive ability 1. Participants completed the Vocabulary and
Block Design subtests of the Wechsler Adult Intelligence Scale-
Revised (WAIS-R; Wechsler, 1981). Because these two subtests
attain moderate (Block Design) and high (Vocabulary) correlations
with the full-scale IQ score, their scores can be prorated to obtain a
popular short-form IQ score with high reliability and validity
(Sattler, 1988). Using Sattler's (1988) formulas, we derived a
 BELIEF BIAS
prorated IQ score for each participant, which we termed general
cognitive ability 1 (CA1). The mean of this variable in the sample
was 108.3 (SD = 11.7).
Cognitive ability 2. The second cognitive ability measure,
cognitive ability 2 (CA2), was also derived by combining perfor-
mance on a verbal and nonverbal test. The latter consisted of 18
problems from Raven's Advanced Progressive Matrices (Set II;
Raven, 1962). The participants were given 15 min to complete the
18 items on the test. By eliminating 12 of the easiest problems,
where performance in a university sample is near ceiling (Carpen-
ter, Just, & Shell, 1990; Raven, Court, & Raven, 1977) and six of
the most difficult problems where performance is nearly floored
(Carpenter et al., 1990; Raven et al., 1977), a cut-time version of
the advanced matrices was created with nearly the same reliability
as the full test (Cahan & Cohen, 1989; Stanovich & Cunningham,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
1992).
This document is copyrighted by the American Psychological Association or one of its allied publishers.
The verbal measure was a brief vocabulary measure (again,
because vocabulary is the strongest specific correlate of general
intelligence; see Matarazzo, 1972). This task employed the checklist-
with-foils format that has been shown to be a reliable and valid way
of assessing individual differences in vocabulary knowledge (R. C.
Anderson & Freebody, 1983; Stanovich, West, & Harrison, 1995).
The stimuli for the task were 40 words and 20 pronounceable
nonwords taken largely from the stimulus list of Zimmerman,
Broder, Shaughnessy, & Underwood (1977). The words and
nonwords were intermixed according to alphabetical order. The
participants were told that some of the letter strings were actual
words and that others were not, and that their task was to read
through the list of items and to put a check mark next to those that
they knew were words. Scoring on the task was determined by
taking the proportion of the target items that were checked and
subtracting the proportion of foils checked.
The CA2 index was created by standardizing the raw scores on
the Raven matrices and the vocabulary checklist measure and then
adding the two standard scores together. The correlation between
the two composite indices, CA1 and CA2, was .707.
Actively Open-Minded Thinking Questionnaire
Participants completed a questionnaire consisting of a number of
subscales assessing styles of epistemic regulation. The following
response format was used for each item in the questionnaire: 6 =
agree strongly, 5 = agree moderately, 4 = agree slightly, 3 =
disagree slightly, 2 — disagree moderately, and 1 = disagree
strongly. The items from the subscales were randomly intermixed,
both with each other and with other scales that were not part of the
present investigation. Most of the subscales were described in more
detail in Stanovich and West (1997). Brief descriptions follow.
Flexible thinking scale. We devised the items on the flexible
thinking scale, which were validated in a previous investigation
(Stanovich & West, 1997). Items tap flexible thinking as a
multifaceted construct that encompasses the cultivation of reflec-
tiveness rather than impulsivity (e.g., "If I think longer about a
problem I will be more likely to solve it"), the seeking and
processing of information that disconfirms one's belief (e.g.,
"People should always take into consideration evidence that goes
against their beliefs"), and the willingness to change one's beliefs
in the face of contradictory evidence (see Baron, 1993, 1994).
There were 10 items on the scale. Because this subscale is
relatively new, the items are reproduced in Appendix A.
Openness-ideas. The eight items from the openness-ideas
facet of the Revised NEO Personality Inventory (Costa & McCrae,
1992) were administered (e.g., "I have a lot of intellectual
curiosity," and "I find philosophical arguments boring"—the latter
was reverse scored).
501
Openness-values. The eight items from the openness-values
facet of the Revised NEO Personality Inventory were administered
(e.g., "I believe that laws and social policies should change to
reflect the needs of a changing world," and "I believe letting
students hear controversial speakers can only confuse and mislead
them"—the latter was reverse scored).
Absolutism. This scale consisted of nine items adapted from
the Scale of Intellectual Development (SID), which was developed
by Erwin (1981,1983). The SID represents an attempt to develop a
multiple-choice scale to measure the early stages of Perry's (1970)
model of intellectual development in young adulthood, which are
characterized by cognitive rigidity, by a belief that issues can be
couched in either-or terms, that there is one right answer to every
complex problem, and by reliance on authority for belief justifica-
tion (e.g., "It is better to simply believe in a religion than to be
confused by doubts about it").
Dogmatism. The dogmatism subscale consisted of nine items.
Three were taken from a short-form field version (Troldahl &
Powell, 1965) of Rokeach's (1960) dogmatism scale, two from
Paulhus and Reid (1991), and four from the full Rokeach scale
published in Robinson, Shaver, and Wrightsman (1991). A typical
item is, "No one can talk me out of something I know is right."
Categorical thinking. The following three items from the
categorical thinking subscale of Epstein and Meier's (1989)
constructive thinking inventory were administered: "There are
basically two kinds of people in this world, good and bad," "I think
there are many wrong ways, but only one right way, to almost
anything," and "I tend to classify people as either for me or against
me."
Belief identification. The belief identification scale was in-
spired by a theoretical paper by Cederblom (1989) in which he
argued for a potential thinking style centered around the extent to
which people identify their beliefs with their self-concept (e.g., "It
makes me happy and proud when someone famous holds the same
beliefs that I do"). A nine-item scale was distilled from Ceder-
blom's discussion of this concept. Because this subscale is
relatively new, the items are reproduced in Appendix A.
Need for closure. Twelve items taken from the Need for
Closure Scale published in Kruglanski, Webster, and Klem (1993)
were administered. A typical item is "I dislike it when a person's
statement could mean many different things."
Procedure
Participants completed the tasks during a single 3-4 hr
session in which they also completed some other tasks that
were not part of the present investigation. All participants
were individually tested by the same experimenter. The
order of tasks completed was as follows: actively open-
minded thinking questionnaire, syllogisms, height judgment
task, WAIS-R subtests, AET, vocabulary checklist, covaria-
tion data evaluation, and Raven matrices.
Results
Projection of the Gender Cue in
the Height Judgment Task
One index of the potency of the gender cue in the
judgments of each participant is the correlation between the
gender of the target in the photograph (G) and that partici-
pant's estimates, r(G, E). The mean value of this correlation
in the ecological set was .693 (SD = .129), and the mean
 502 SA, WEST, AND STANOVICH
value of r(G,E) in the matched set was .281 (SD = .196)—a
difference that was statistically significant, r(116) = 26.37,
p < .001. The actual correlation between gender and the
target heights in the photographs in the ecological set, r(G,
T), was .628. Thus, the mean r(G, E) value of .693 in the
ecological set represents only slightly more projection of the
gender cue than is actually warranted. In the matched
condition, the r(G, T) correlation was deliberately con-
strained to zero (actually, .018), therefore the mean r(G, E)
value of .281, which is significantly greater than zero,
f(116) = 15.48,/? < .001, reliably represents more projec-
tion of the gender cue in this condition than is warranted.
Another metric that captures the degree of projection of
the gender cue is simply the mean difference between the
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
estimates of male targets and female targets for each
This document is copyrighted by the American Psychological Association or one of its allied publishers.
participant. The mean difference in the ecological condition,
(M - FOecoiogica], ranged from 1.53 to 9.37 in. and averaged
4.13 in. (SD — 1.25). This mean is almost one inch less than
the actual mean difference in the ecological set (5 in.). The
mean difference in the matched condition, (M — F)matched,
was substantially lower (1.21, SD = 0.91), but it was still
significantly different from zero, f(116) = 14.35, p < .001.
In the matched condition, despite the instructions emphasiz-
ing that the males and females were matched in height, 111
of 117 participants gave mean male height estimates that
were larger than their female estimates.
Accuracy in the Height Judgment Task: The Efficacy
and Inefficacy of Projecting Prior Belief
There are two indices of accuracy in the height judgment
task, each reflecting a different property of estimation
efficacy (see N. R. Brown & Siegler, 1993). One index is the
correlation between the estimates (E) and the target heights
of the actual models (T). This correlation, r(T, E), was
calculated for each participant and ranged from .395 to .766
in the ecological set and from .091 to .645 in the matched
set. The mean value of r(T, E) was considerably higher in the
ecological set (.610, SD = .071) than in the matched set
(.358, SD = .109), r(116) = 20.34, p < .001.
The r(T, E) accuracy index reflects the tendency to order
the targets correctly but it does not reflect the tendency to
use the scale correctly—that is, to arrive at estimates that
actually match the target heights. The sum of the absolute
deviations (SAD) between estimate and target across all
stimuli reflects this property. The smaller this index, the
higher the absolute accuracy of the estimates. In principle,
these two indices can be independent. For example, it is
possible for r(T, E) to be quite high (indicating highly
accurate ordering), and for SAD to also be high (indicating
poor absolute accuracy of the estimates, i.e., poor scale use).
This might occur if someone ordered the photographs nearly
perfectly but had each of the estimates 3-4 inches too tall
(because the individual believed that, overall, the population
is taller than it is).
We might expect that in the ecological set, projecting
one's knowledge of the gender relationship would be
efficacious because in that set of stimuli there actually is a
correlation between gender and target height, r(G, T) =
.628. We found support for this conjecture. Using r(G, E) as
a measure of the projection of this prior belief, we found that
this index correlated highly with both r(T, E), r = .697, p <
.001, and SAD, r - -.525, p < .001. That is, the more the
participants projected the gender-height relationship, the
more accurate they were. Participants who more strongly
projected the gender-height relationship in their estimates
tended to order the stimuli more accurately, and they made
estimates that tracked the actual heights more closely in an
absolute sense.
In the ecological set, even the tendency to overuse the
gender cue (in short, to stereotype) is efficacious. This point
is made clear in a subsequent analysis. Indicated in the top
half of Figure 1 (analogous to the figures in Nelson et al.,
1990), is the outcome when the aggregate height estimates
(averaged over all 117 respondents) are regressed on gender
and the actual height of the targets (additional regression
analyses were run with image size as an additional predictor,
but this variable had virtually no effect on this or any
subsequent analysis, primarily because it had negligible
correlations with gender, .005, actual height, —.159, and
estimated height, -.018). The highly significant standard-
Ecological Height
Judgment Task
Height
Estimate
0.610" >' >v0.343"
Actual
Gender
Height
0.628"
Matched Height
Judgment Task
Height
Estimate
0.388* >* \ . 0.500"
Actual
Gender
Height
0.018
Figure I. Top: Path diagram predicting the aggregate height
judgments in the ecological set. Bottom: Path diagram predicting
the aggregate height judgments in the matched set. Numbers on
arrows are standardized beta weights.
*p < .01. **p < .001.
 BELIEF BIAS
ized beta weight for gender is an indication that, in the words
of Nelson et al. (1990), "male targets were judged to be
taller, on average, than the female targets, even after the
actual difference in height between the male and female
targets had been statistically controlled" (p. 665). However,
the tendency to stereotype on the basis of gender is actually
associated with more accurate estimates. This analysis was
carried out by running the regression analysis pictured in the
top half of Figure 1 on each individual participant's scores
and estimating 117 separate beta weights for gender and
actual height (one pair for each participant). The former
parameter was positively correlated with r(T, E), r = .443,
p < .001, and negatively correlated with SAD, r = -.314,
p < .001. Thus, the more individuals judged male targets to
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
be taller, even after actual differences in male and female
This document is copyrighted by the American Psychological Association or one of its allied publishers.
targets were statistically controlled, the more accurate they
were, both in correctly ordering the stimuli and in their
absolute deviations from the actual heights.
Of course, in the matched set, projection of the gender cue
is not efficacious. There, the projection index, r(G, E), did
not correlate significantly with either r(T, E), r = -.172, or
SAD, r = .143—the more the participants projected the
gender-height relationship, the less accurate they were. In
the analysis that is summarized in the bottom half of Figure
1, aggregate height estimates (averaged over all 117 respon-
dents) were regressed on gender and the actual height of the
targets in the matched set. The standardized beta weight for
gender—although reduced in magnitude from that obtained
with the ecological set—was highly significant. Despite the
instructions emphasizing that these targets were matched for
height across gender, participants persisted in estimating
greater male heights. Here, unlike the case in the ecological
set, the tendency to stereotype on the basis of gender was not
associated with more accurate estimates in the matched set.
The regression analysis pictured in Figure 1 was run on each
individual participant, and 117 separate beta weights for
gender and actual height (one pair for each participant) were
estimated. The former parameter failed to correlate posi-
tively with r(T, E), r = -.181, / ? < .05, or negatively with
SAD, r = .148, ns. In fact, the signs of the correlations were
in the opposite direction, significantly so in the former
case—more stereotyped responders ordered the photographs
less accurately.
Performance on the Syllogistic Reasoning Task
The mean number of correct responses on the syllogistic
reasoning task was 19.2 (SD = 4.2). The mean number of
consistent items answered correctly was 7.02 (SD = 1.17),
the mean number of neutral items answered correctly was
6.71 (SD = 1.58), and the mean number of inconsistent
items answered correctly was 5.49 (SD = 2.05). There were
indications of a significant belief bias effect (i.e., difference
from the neutral condition) on both the consistent items,
r(123) = 3.37, p < .01, and the inconsistent items, f(123) =
8.64, p < .001, although the effect was larger for the latter.
As an overall index of belief bias, we employed the
difference score, which was obtained by subtracting the
number correct on the inconsistent items from the number
503
correct on the consistent items. This belief bias index could,
in principle, range from - 8 to +8, but the actual range was
from — 1 to 8. Positive scores indicate some degree of belief
bias, with the higher scores indicating more belief bias. The
mean belief bias score was 1.53 (SD = 1.77), and it was
significantly different from zero, f(123) = 9.62, p < .001; 82
of 124 participants displayed some degree of belief bias.
Is Projection of Prior Belief in Height Judgment
Associated With Belief Bias in a Verbal Task?
We now turn to the question of whether there is any
domain generality to the projection of prior belief across
verbal and nonverbal tasks. Several correlations in the
matrix displayed in Table 1 are relevant to this question. The
first two variables listed in Table 1 reflect the degree of pro-
jection of the gender relationship onto the estimates, r(G, E),
in the matched and ecological sets, respectively. The degree
of belief bias on the syllogistic reasoning task was signifi-
cantly correlated with the former but not with the latter (see
the fourth row of Table 1). The correlations were different in
sign and they were significantly different from each other:
.209 versus -.095, f(114) = 3.52, p < .001, test for
difference in dependent correlations (see Cohen & Cohen,
1983, pp. 56-57). Thus, it is the condition where the
participant is instructed to ignore the relationship between
gender and height that correlates with belief bias in the
syllogisms task and not the ecological condition where the
gender relationship is in fact predictive.
We calculated another index from the height judgment
task that reflects the ability to ignore prior knowledge and to
follow the instructions in the matched set, which instructed
the participants to not use gender as a height judgment cue.
The need for this index derives from a property shared by
both pure projection indices in the matched condition. The
pure projection indices from the matched condition, r(G,
E)matched and (M - F)matched, should both be zero if the
participant fully responded to the instructions and removed
all influence of the prior belief—any deviation from the zero
value is interpreted as an intrusion of prior knowledge into
the height judgments. The problem comes about because
reducing these indices to zero may be differentially difficult,
not because of variation in the belief bias itself, but because
the strength of the gender stereotype may vary from
participant to participant. To put it simply, it should be easier
for a participant who previously thought that there was on
average, a 2-in. difference between males and females to
reduce (M — F)matched to zero than it would be for a
participant who thought that there was a 7-in. difference
between males and females. The prior belief of the latter
participant is much more strongly contradicted by the
matched set. Fortunately, in the ecological set, we actually
have a measure of the participant's prior belief about the
strength of the gender-height relationship, (M - F)ecoiogical.
We used this estimate of the prior belief to construct an index
of the degree of adjustment from the ecological set that was
achieved in the matched set. This index, HJ Adjustment
(height judgment adjustment), was simply (M - F ^ o ^ a ,
minus (M - F)matched—the mean difference between male
 504 SA, WEST, AND STANOVICH

Table 1
Intercorrelations Among the Primary Variables
Variable 1
Belief bias indices
1. r(G, EJmatched —
2. KG.EUiogica, .526 —
3. HJ Adjustment Score -.396 .342 —
4. Syllogisms, Belief Bias .209 -.095 -.341 —
Cognitive decontextualization tasks
5. Covariation Judgment -.200 .168 .322 -.342 —
6. AET -.261 .101 .284 -.236 .324 —
Cognitive ability measures
7. CA1 -.155 .256 .347 -.445 .493 .449 —
8. CA2 -.215 .263 .461 -.495 .505 .392 .707 —
Styles ofepistemic regulation
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

9. AOT -.130 .057 .119 -.315 .241 .436 .358 .323 —

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Note. Correlations larger than .182 are significant at the .05 level (two-tailed); correlations larger
than .237 are significant at the .01 level (two-tailed); and correlations larger than .299 are significant
at the .001 level (two-tailed). KG. E)matched = correlation between gender and estimates in the
matched condition; r(G, E^iogjcai = correlation between gender and estimates in the ecological
condition; HJ Adjustment Score = (M - F)ecoiogicai minus (M - F)ma,ched; CA1 = cognitive ability
composite 1 (WAIS block design and WAIS vocabulary); CA2 = cognitive ability composite 2
(Raven Matrices and vocabulary); AET = argument evaluation test; AOT = composite actively
open-minded thinking scale.

and female height estimates in the ecological set minus the
mean difference in the matched set. This index captures how
much the participant succeeded in reducing his or her
estimates of male-female difference (use of a proportional
score rather than a difference score produced very similar
results). The mean HJ Adjustment score was 2.92
(SD = 1.2)—the average participant decreased the male-
female difference in their estimates by almost 3 in. The
range in this index was quite large, from .03 in. to 7.41 in.
Presented in the third column of Table 1 are the correla-
tions involving the HJ Adjustment index. Its correlation with
belief bias on the syllogistic reasoning task was even larger
in absolute magnitude than the simpler belief projection
index, r(G, E)matchKl (-.341 vs. .209; the correlation with the
former is negative because those showing larger adjustments
to the matched set instructions displayed less belief bias in
syllogistic reasoning).
Is Projection of Prior Belief Associated With
Cognitive Ability and Cognitive Decontextualization?
The two composite indices of general cognitive ability,
CA1 and CA2, were both negatively correlated with projec-
tion in the matched set and positively correlated with
projection in the ecological set. In both cases, the correla-
tions with r(G, E)matched and r(G, E)ecological were significantly
different, f(114) = 4.98, p < .001, and f(114) = 6.02, p <
.001, in the cases of CA1 and CA2, respectively. Consistent
with these findings are the significant moderate correlations
with HJ Adjustment displayed by CA1 and CA2 (.347 and
.461, respectively).
Similar patterns were apparent for the two cognitive
decontextualization tasks—the covariation judgment task
and the AET. They were both significantly correlated with
belief projection in the matched set but not with the belief
projection in the ecological set. In both cases, the signs of
the pairs of correlations were different, and each pair of
correlations was significantly different from each other,
f(114) = 4.30, p < .001, andf(113) = 4.34, p < .001, for the
AET and covariation task, respectively. Significant correla-
tions were observed between the HJ Adjustment score and
the covariation judgment task (.322) and the HJ Adjustment
score and the AET (.284).
In summary, cognitive ability and decontextualization
skill was positively correlated with belief bias in a percep-
tual judgment task where prior belief accurately reflected an
aspect of the perceptual environment and negatively corre-
lated with belief bias in a perceptual judgment task where
prior belief was incongruent with the perceptual environ-
ment. This finding is consistent with the notion of human
intelligence as adaptive to context (Steinberg, 1985, 1997).
Given the findings on the height judgment task, it then
becomes interesting to examine whether cognitive ability
and decontextualization skill are associated with belief bias
on a verbal reasoning task. Both indices of cognitive ability,
CA1 and CA2, displayed significant and moderate negative
correlations with belief bias on the syllogistic reasoning task
(-.445 and —.495), as did performance on the covariation
task (-.342) and the AET (-.236). The direction of all of
these correlations was consistent: Individuals higher in
cognitive ability displayed less belief bias, and they scored
higher on the two measures of the ability to reason indepen-
dently of prior belief (the AET and the covariation judgment
task).
Relationships With a Composite Actively
Open-Minded Thinking Score
Because the eight subscales on the actively open-minded
thinking questionnaire displayed moderate intercorrelations
 BELIEF BIAS
Table 2
Hierarchical Regression Analyses
Step Variable R
^change
Predicting belief bias in syllogistic reasoning
1 CA1 .445 .198
2 AOT .476 .028
Predicting argument evaluation test performance
1 CA1 .449 .202
2 AOT .537 .086
Predicting covariation judgment performance
1 CA1 .493 .243 38.81**
2 AOT .497 .004
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
Note. CA1 = cognitive ability composite 1 (WAIS block design and WAIS vocabulary); AOT •
composite actively open-minded thinking scale.
*/7<.O5. **p<. 01.
(see Appendix B, Table Bl), a composite actively open-
minded thinking (AOT) score was formed. First, the total
scores on each of the subscales were standardized. Then, the
standard scores on the flexible thinking, openness-ideas,
and openness-values subscales were summed. From this
sum was subtracted the sum of the standard scores on the
absolutism, dogmatism, categorical thinking, belief identifi-
cation, and need for closure subscales (the creation of a
composite score using the factor score from the first factor
produced results virtually identical to those from the unit-
weighted sum of standard scores). Thus, high scores on the
AOT composite indicate openness to belief change and
cognitive flexibility, whereas low scores indicate cognitive
rigidity and resistance to belief change. The creation of the
composite score was justified by a factor analysis conducted
on the eight subscales (see Appendix B, Table B2). Only one
factor had an eigenvalue greater than one. All variables had
loadings greater than .300 on this factor. This factor
accounted for 38.7% of the variance, whereas the second
factor accounted for only 6.1 % of the variance.
The actively open-minded thinking composite score did
not significantly correlate with projecting the gender relation-
ship in the height judgment task, but it did display significant
correlations with belief bias on the syllogistic reasoning task
(see last line of Table 1). This result is consistent with
previous demonstrations that measures of epistemic regula-
tion can predict belief bias in verbal tasks (Kardash &
Scholes, 1996; Klaczynski et al., 1997). In fact, the actively
open-minded thinking composite score predicted belief bias
in the syllogistic reasoning task even after differences in
cognitive ability were partialled out. A multiple R of .476
was observed when the actively open-minded thinking
composite score and the prorated WAIS IQ score (CA1;
similar results were obtained with CA2) were entered as
predictors of belief bias in syllogistic reasoning (see the first
hierarchical multiple regression analysis in Table 2). As
indicated in Table 2, CA1 was a significant unique predictor
(partial correlation = -.375), but so was the actively open-
minded thinking composite score (partial correla-
tion = -.186). This trend was even more apparent in the
505
'change Final F Partial r
30.21** 19.86** -.375
4.36* 4.36* -.186
30.80** 16.76** .349
14.77** 14.77** .330
30.10** .349
0.66 0.66 .074
next analysis in Table 2, where performance on the AET was
the criterion variable. There, the actively open-minded
thinking composite score accounted for 8.6% unique vari-
ance after CA1 was entered into the regression equation
{p < .01), and its partial correlation was almost as large as
that of the cognitive ability measure (.330 vs. .349).
However, a different outcome was obtained when using
covariation judgment performance as the criterion variable.
There, despite a significant zero-order correlation (see Table
1), the AOT composite score did not account for significant
variance after cognitive ability was entered into the regres-
sion equation.
General Discussion
In response to the relatively modest reduction in the
gender-height stereotype that they achieved with their
matched set manipulation, Nelson et al. (1990) commented
that "people may be largely unable to control the influence
of real-life base rates (e.g., the stimulus-response associa-
tion between sex and height) that have been built up over a
lifetime of experience, despite their best attempts to do so"
(p. 672). We observed a somewhat more substantial reduc-
tion (from a mean height difference of 4.13 in. to one of 1.21
in.), but more importantly, we have documented very large
individual differences in die ability to "control the influence
of real-life base rates" and have shown that the ability to
control the influence of prior knowledge in this perceptual
judgment task is related to the ability to avoid belief bias on
a verbal reasoning task. Scores on two different measures of
cognitive ability (CA1 and CA2) and two different measures
of cognitive decontextualization (AET and covariation judg-
ment task) were also related to the ability to separate height
judgments from the gender stereotype. Thus, people can
"control the influence of real-life base rates," but there are
large individual differences in the ability to do so—
differences that are in part predictable from other reasoning
and problem solving abilities.
That individual differences in the ability to avoid gender
projection in the matched set is predictable is related to the
 506 SA, WEST, AND STANOVICH
issue of the domain specificity and generality raised in the
introduction. An extreme form of domain specificity—one
where the ability to evaluate evidence in an unbiased manner
is completely independent from domain to domain—was
clearly falsified by the results presented here. The ability to
separate the influence of prior knowledge from the estimate
required in the height judgment task (a perceptual judgment
situation) was linked to the ability to separate prior knowl-
edge from judgments of logical validity in the syllogistic
reasoning task (a purely verbal task).
Although an extreme form of the domain specificity view
is falsified by these results, the results regarding the actively
open-minded style of epistemic regulation were somewhat
more supportive. There, it was found that the relationships
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
with a verbal reasoning measure of belief bias (the syllogis-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
tic reasoning task) were quite different from those involving
belief bias in the height judgment task. In the latter, there
were no significant zero-order correlations with the actively
open-minded thinking composite score. However, in the
former, not only was the zero-order correlation significant,
but the actively open-minded thinking composite score
accounted for variance in the belief bias effect even after the
variance associated with a measure of cognitive ability had
been partialled out.
The actively open-minded thinking (AOT) composite
score was also significantly correlated with performance on
both of the cognitive decontextualization tasks: the AET and
the covariation judgment task. However, the AOT composite
score was a unique predictor (after cognitive ability was
partialled out) of the former but not the latter (see Table 2).
The difference in the importance of epistemic regulation
across these two tasks might relate to a distinction intro-
duced by Wilson and Brekke (1994) in their classification of
reasoning errors. They distinguished reasoning errors involv-
ing the failure to apply an appropriate rule from those
occurring in situations where there exist no appropriate rule
or algorithm that guarantees problem solution. The second
situation is conducive to what Wilson and Brekke termed
mental contamination—unwanted influences on judgments
and evaluations. Whereas an incorrect evaluation of the data
in the covariation judgment task reflects the failure to
approximate a rule (the Ap rule of conditional probability),
there is no clear-cut rule or algorithm to apply to the
argument-evaluation items. Thus, the AET may be suscep-
tible to mental contamination from prior beliefs in a manner
not as obvious in the covariation judgment task. Epistemic
regulation in the context of the AET might consist of
metacognitive recognition of the lack of definitive response
rules, followed by attempts to control the influence of the
likely contaminating effects of prior belief. Our results from
the AOT scale demonstrated that indirect indicators of
tendencies toward such epistemic regulation predicted per-
formance on the AET over and above the variance accounted
for by cognitive ability.
The results involving the cognitive ability measures were
very much in line with an adaptive interpretation of the
nature of human intelligence (Larrick, Nisbett, & Morgan,
1993; Steraberg, 1985, 1997; Sternberg & Detterman,
1986). Both cognitive ability measures (CA1 and CA2) were
positively correlated with belief bias in a perceptual judg-
ment task where prior knowledge accurately reflected an
aspect of the perceptual environment (the ecological set) and
negatively correlated with belief bias in a perceptual judg-
ment task where prior knowledge was incongruent with the
perceptual environment (the matched set)—just the pattern
that an adaptive view of intelligence would predict.
Our results on styles of epistemic regulation are consistent
with those of Klaczynski et al. (1997), but our results with
respect to cognitive ability are not. They found no correla-
tion between cognitive ability and belief bias in evaluating
arguments and evaluating evidence, whereas we found
significant correlations on our syllogistic reasoning measure.
Aside from some clear differences in the tasks, we think one
difference might be that our participants, unlike theirs, were
explicitly and specifically instructed to ignore prior knowl-
edge and belief in all our tasks. Thus, our situations might
have been more prone to implicate metacognitive control
processes that are strongly identified with intelligence
(Byrnes, 1995; Sternberg, 1985).
Cognitive ability was associated with skill at detaching
prior knowledge from reasoning in the syllogistic reasoning
task. If contextualist theories were correct, and tendencies
toward decontextualization are actually maladaptive, then
people with increased cognitive capacity might be expected
to display greater reluctance to detach world knowledge
from their information processing. Thus, a strong version of
the contextualist position—that, because of the primacy of
context-based processing (Hilton, 1995; Levinson, 1995;
Stanovich, 1999), more cognitively able individuals would
be more likely to carry over contextual information into an
environment where context was inefficacious—was not
supported. Our results concerning cognitive ability thus
seem to bolster the normative status of styles of epistemic
regulation that emphasize unbiased reasoning (styles that are
focal to the critical thinking literature), but with an important
caveat that is consistent with the contextualist tradition.
More intelligent individuals do contextualize the problem
more when that context contains cues that can facilitate
judgment, but they are less likely to carry over contextual
cues into situations where they know the cues are no longer
diagnostic.
Generating at least contingent support for the normative
tradition in the critical thinking literature is important
because several philosophical analyses have called into
question the normative status of the stricture that evidence
evaluation not be contaminated by prior belief. For example,
Kornblith (1993) argued:
Mistaken beliefs will, as a result of belief perseverance, taint
our perception of new data. By the same token, however,
belief perseverance will serve to color our perception of new
data when our preexisting beliefs are accurate . . . If, overall,
our belief-generating mechanisms give us a fairly accurate
picture of the world, then the phenomenon of belief persever-
ance may do more to inform our understanding than it does to
distort it. (p. 105)
This argument—that in a natural ecology where most of our
prior beliefs are true, projecting our beliefs onto new data
will lead to faster accumulation of knowledge—has been
 BELIEF BIAS
termed the knowledge projection argument (see Stanovich,
1999), and it reappears in a remarkably diverse set of
contexts throughout the reasoning and decision making
literature (Dawes, 1990; Edwards & Smith, 1996; Evans et
al., 1993; Koehler, 1993; Krueger & Zeiger, 1993; see
Stanovich, 1999, for a review).
The contingent validity of this argument and its interac-
tion with individual differences is perhaps best conceptual-
ized within two-process models of cognition. Specifically,
numerous theorists have proposed two-process models of
cognitive activity that distinguish automatic heuristic process-
ing from controlled analytic processing (e.g., Epstein, 1994;
Evans & Over, 1996; Sloman, 1996; see Stanovich, 1999,
for a review). Instructions requiring decontextualization
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
(such as instructions to ignore a prior belief) probably
heavily stress the analytic system, which must be employed
to override the heuristic system when the latter's contextual-
ized response is inappropriate (Stanovich, 1999). Because
analytic processing is more closely related to computational
capacity than is heuristic processing (Evans & Over, 1996;
McGeorge, Crawford, & Kelly, 1997; Reber, 1993), under
the two-process view it might be expected that successful
decontextualization would be associated with higher cogni-
tive ability.
References
Alexander, P. A., Kulikowich, J. M., & Schulze, S. K. (1994). How
subject-matter knowledge affects recall and interest. American
Educational Research Journal, 31, 313-337.
Allan, L. G. (1980). A note on measurement of contingency
between two binary variables in judgment tasks. Bulletin of the
Psychonomic Society, 15, 147-149.
Anderson, J. R., Reder, L. M., & Simon, H. A. (1996). Situated
learning and education. Educational Researcher, 25(4), 5-11.
Anderson, R. C , & Freebody, P. (1983). Reading comprehension
and the assessment and acquisition of word knowledge. In B.
Huston (Ed.), Advances in reading/language research (Vol. 2,
pp. 231-256). Greenwich, CT: JAI Press.
Baron, J. (1985a). Rationality and intelligence. Cambridge, En-
gland: Cambridge University Press.
Baron, J. (1985b). What kinds of intelligence components are
fundamental? In S. Chipman & J. Segal (Eds.), Thinking and
learning skills (Vol. 2, pp. 365-390). Hillsdale, NJ: Erlbaum.
Baron, J. (1991). Beliefs about thinking. In J. Voss, D. Perkins, & J.
Segal (Eds.), Informal reasoning and education (pp. 169-186).
Hillsdale, NJ: Erlbaum.
Baron, J. (1993). Why teach thinking?—An essay. Applied Psychol-
ogy: An International Review, 42, 191-214.
Baron, J. (1994). Thinking and deciding (2nd ed.). New York:
Cambridge University Press.
Baron, J. (1995). Myside bias in thinking about abortion. Thinking
and Reasoning, 1, 221-235.
Baron, J. B., & Sternberg, R. (Eds.). (1987). Teaching thinking
skills: Theory and practice. New York: Freeman.
Biek, M , Wood, W, & Chaiken, S. (1996). Working knowledge,
cognitive processing, and attitudes: On the determinants of bias.
Personality and Social Psychology Bulletin, 22, 547-556.
Broniarczyk, S., & Alba, J. W. (1994). Theory versus data in
prediction and correlation tasks. Organizational Behavior and
Human Decision Processes, 57, 117-139.
507
Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition
and the culture of learning. Educational Researcher, 18(1),
32-42.
Brown, N. R., & Siegler, R. S. (1993). Metrics and mappings: A
framework for understanding real-world quantitative estimation.
Psychological Review, 100, 511-534.
Byrnes, J. P. (1995). Domain-specificity and the logic of using
general ability as an independent variable or covariate. Merrill-
Palmer Quarterly, 41, 1-24.
Cahan, S., & Cohen, N. (1989). Age versus schooling effects on
intelligence development. Child Development, 60, 1239-1249.
Carpenter, P. A., Just, M. A., & Shell, P. (1990). What one
intelligence test measures: A theoretical account of the process-
ing in the Raven Progressive Matrices Test. Psychological
Review, 97,404-431.
Ceci, S. J. (1993). Contextual trends in intellectual development.
Developmental Review, 13, 403-435.
Ceci, S. J. (1996). On intelligence: A bioecological treatise on
intellectual development (Expanded ed). Cambridge, MA: Har-
vard University Press.
Cederblom, J. (1989). Willingness to reason and the identification
of the self. In E. P. Maimon, B. F. Nodine, & F. W. O'Conner
(Eds.), Thinking, reasoning, and writing (pp. 147-159). New
York: Longman.
Cohen, J., & Cohen, P. (1983). Applied multiple regression/
correlation analysis for the behavioral sciences (2nd ed.).
Hillsdale, NJ: Erlbaum.
Costa, P. T, & McCrae, R. R. (1992). Revised NEO personality
inventory. Odessa, FL: Psychological Assessment Resources.
Dawes, R. M. (1990). The potential nonfalsity of the false
consensus effect. In R. M. Hogarth (Ed.), Insights into decision
making (pp. 179-199). Chicago: University of Chicago Press.
Dole, J. A., & Sinatra, G. M. (1998). Reconceptualizing change in
the cognitive construction of knowledge. Educational Psycholo-
gist, 33, 109-128.
Edwards, K., & Smith, E. E. (1996). A disconfirmation bias in the
evaluation of arguments. Journal of Personality and Social
Psychology, 71, 5-24.
Ennis, R. H. (1987). A taxonomy of critical thinking dispositions
and abilities. In J. Baron & R. Sternberg (Eds.), Teaching
thinking skills: Theory and practice (pp. 9-26). New York:
Freeman.
Epstein, S. (1994). Integration of the cognitive and the psychody-
namic unconscious. American Psychologist, 49, 709-724.
Epstein, S., & Meier, P. (1989). Constructive thinking: A broad
coping variable with specific components. Journal of Personal-
ity and Social Psychology, 57, 332-350.
Erwin, T. D. (1981). Manual for the Scale of Intellectual Develop-
ment. Harrisonburg, VA: Developmental Analytics.
Erwin, T. D. (1983). The scale of intellectual development:
Measuring Perry's scheme. Journal of College Student Person-
nel, 24, 6-12.
Evans, J. S. B. T, Barston, J., & Pollard, P. (1983). On the conflict
between logic and belief in syllogistic reasoning. Memory &
Cognition, 11, 295-306.
Evans, J. S. B. T, & Over, D. E. (1996). Rationality and reasoning.
Hove, England: Psychology Press.
Evans, J. S. B. T , Over, D. E., & Manktelow, K. (1993). Reasoning,
decision making and rationality. Cognition, 49, 165-187.
George, C. (1995). The endorsement of the premises: Assumption-
based or belief-based reasoning. British Journal of Psychology,
56,93-111.
Hilton, D. J. (1995). The social context of reasoning: Conversa-
tional inference and rational judgment. Psychological Bulletin,
118, 248-271.
 508 SA, WEST, AND STANOVICH
Kao, S. R, & Wasserman, E. A. (1993). Assessment of an
information integration account of contingency judgment with
examination of participative cell importance and method of
information presentation. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 19, 1363—1386.
Kardash, C. M., & Scholes, R. J. (1996). Effects of pre-existing
beliefs, epistemological beliefs, and need for cognition on
interpretation of controversial issues. Journal of Educational
Psychology, 88, 260-271.
Klaczynski, P. A. (1997). Bias in adolescents' everyday reasoning
and its relationship with intellectual ability, personal theories,
and self-serving motivation. Developmental Psychology, 33,
273-283.
Klaczynski, P. A., & Gordon, D. H. (1996). Self-serving influences
on adolescents' evaluations of belief-relevant evidence. Journal
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
of Experimental Child Psychology, 62, 317-339.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
Klaczynski, P. A., Gordon, D. H., & Fauth, J. (1997). Goal-oriented
critical reasoning and individual differences in critical reasoning
biases. Journal of Educational Psychology, 89, 470-485.
Klaczynski, P. A., & Narasimham, G. (1998). Development of
scientific reasoning biases: Cognitive versus ego-protective
explanations. Developmental Psychology, 34, 175-187.
Koehler, J. J. (1993). The influence of prior beliefs on scientific
judgments of evidence quality. Organizational Behavior and
Human Decision Processes, 56, 28-55.
Kornblith, H. (1993). Inductive inference and its natural ground.
Cambridge, MA: MIT Press.
Krueger, J., & Zeiger, J. (1993). Social categorization and the truly
false consensus effect. Journal of Personality and Social Psychol-
ogy, 65, 670-680.
Kruglanski, A. W., Webster, D. M., & Klem, A. (1993). Motivated
resistance and openness to persuasion in the presence or absence
of prior information. Journal of Personality and Social Psychol-
ogy, 65, 861-876.
Kuhn, D. (1991). The skills of argument. Cambridge, England:
Cambridge University Press.
Kuhn, D. (1996). Is good thinking scientific thinking? In D. R.
Olson & N. Torrance (Eds.), Modes of thought: Explorations in
culture and cognition (pp. 261-281). New York: Cambridge
University Press.
Kunda, Z. (1990). The case for motivated reasoning. Psychological
Bulletin, 108, 480-498.
Larrick, R. P., Nisbett, R. E., & Morgan, J. N. (1993). Who uses the
cost-benefit rules of choice? Implications for the normative
status of microeconomic theory. Organizational Behavior and
Human Decision Processes, 56, 331-347.
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate
peripheral participation. Cambridge, England: Cambridge Uni-
versity Press.
Levin, I. P., Wasserman, E. A., & Kao, S. F. (1993). Multiple
methods of examining biased information use in contingency
judgments. Organizational Behavior and Human Decision Pro-
cesses, 55, 228-250.
Levinson, S. C. (1995). Interactional biases in human thinking. In
E. Goody (Ed.), Social intelligence and interaction (pp. 221-
260). Cambridge, England: Cambridge University Press.
Lipman, M. (1991). Thinking in education. Cambridge, England:
Cambridge University Press.
Lord, C. G., Lepper, M. R., & Preston, E. (1984). Considering the
opposite: A corrective strategy for social judgment. Journal of
Personality and Social Psychology, 47, 1231-1243.
Markovits, H., & Nantel, G. (1989). The belief-bias effect in the
production and evaluation of logical conclusions. Memory &
Cognition, 17, 11-17.
Matarazzo, J. D. (1972). Wechsler's measurement and appraisal of
adult intelligence (5th ed.). Baltimore: Williams & Wilkins.
McGeorge, P., Crawford, J., & Kelly, S. (1997). The relationships
between psychometric intelligence and learning in an explicit
and an implicit task. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 23, 239—245.
Moshman, D. (1994). Reasoning, metareasoning, and the promo-
tion of rationality. In A. Demetriou & A. Efklides (Eds.),
Intelligence, mind, and reasoning: Structure and development
(pp. 135-150). Amsterdam: Elsevier.
Moshman, D., & Franks, B. (1986). Development of the concept of
inferential validity. Child Development, 57, 153-165.
National Dairy Council. (1980). Recommended dietary allowances
revised (NASNRC Publication No. ISSN 0111-5568). Rose-
mont, IL: Author.
Nelson, T., Biernat, M., & Manis, M. (1990). Everyday base rates
(sex stereotypes): Potent and resilient. Journal of Personality
and Social Psychology, 59, 664-675.
Nickerson, R. S. (1987). Why teach thinking? In J. Baron & R.
Sternberg (Eds.), Teaching thinking skills: Theory and practice
(pp. 27-40). New York: Freeman.
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenom-
enon in many guises. Review of General Psychology, 2, 175-
220.
Norris, S. P., & Ennis, R. H. (1989). Evaluating critical thinking.
Pacific Grove, CA: Midwest Publications.
Paulhus, D. L., & Reid, D. B. (1991). Enhancement and denial in
socially desirable responding. Journal of Personality and Social
Psychology, 60, 307-317.
Perkins, D. N. (1995). Outsmarting IQ: The emerging science of
learnable intelligence. New York: Free Press.
Perkins, D. N., Jay, E., & Tishman, S. (1993). Beyond abilities: A
dispositional theory of thinking. Merrill-Palmer Quarterly, 39,
1-21.
Perry, W. G. (1970). Forms of intellectual and ethical development
in the college years: A scheme. New York: Holt, Rinehart &
Winston.
Raven, J. C. (1962). Advanced Progressive Matrices (Set II).
London: H. K. Lewis.
Raven, J. C , Court, J. H., & Raven, J. (1977). Manual for
Advanced Progressive Matrices (Sets I & II). London: H. K.
Lewis.
Reber, A. S. (1993). Implicit learning and tacit knowledge. New
York: Oxford University Press.
Robinson, J. P., Shaver, P. R., & Wrightsman, L. S. (1991).
Measures ofpersonality and social psychological attitudes (Vol.
1, pp. 560-564). San Diego, CA: Academic Press.
Rogoff, B., & Lave, J. (Eds.). (1984). Everyday cognition. Cam-
bridge, MA: Harvard University Press.
Rokeach, M. (1960). The open and closed mind. New York: Basic
Books.
Sattler, J. M. (1988). The assessment of children (3rd ed.). San
Diego: Jerome Sattler Publishers.
Schommer, M. (1990). Effects of beliefs about the nature of
knowledge on comprehension. Journal of Educational Psychol-
ogy, 82, 498-504.
Schommer, M. (1993). Epistemological development and aca-
demic performance among secondary students. Journal of
Educational Psychology, 85, 406—411.
Schommer, M. (1994). Synthesizing epistemological belief re-
search: Tentative understandings and provocative confusions.
Educational Psychology Review, 6, 293-319.
Schommer, M., & Walker, K. (1995). Are epistemological beliefs
similar across domains? Journal of Educational Psychology, 87,
424-^32.
 BELIEF BIAS
Schustack, M. W., & Stemberg, R. J. (1981). Evaluation of
evidence in causal inference. Journal of Experimental Psychol-
ogy: General, 110, 101-120.
Shanks, D. R. (1995). Is human learning rational? Quarterly
Journal of Experimental Psychology, 48A, 257-279.
Siegel, H. (1988). Educating reason. New York: Routledge.
Siegel, H. (1997). Rationality redeemed? Further dialogues on an
educational ideal. New York: Routledge.
Sloman, S. A. (1996). The empirical case for two systems of
reasoning. Psychological Bulletin, 119, 3-22.
Slusher, M. P., & Anderson, C. A. (1996). Using causal persuasive
arguments to change beliefs and teach new information: The
mediating role of explanation availability and evaluation bias in
the acceptance of knowledge. Journal of Educational Psychol-
ogy, 88, 110-122.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Stanovich, K. E. (1999). Who is rational? Studies of individual
This document is copyrighted by the American Psychological Association or one of its allied publishers.
differences in reasoning. Mahwah, NJ: Erlbaum.
Stanovich, K. E., & Cunningham, A. E. (1992). Studying the
consequences of literacy within a literate society: The cognitive
correlates of print exposure. Memory & Cognition, 20, 51-68.
Stanovich, K. E., & West, R. F. (1997). Reasoning independently of
prior belief and individual differences in actively open-minded
thinking. Journal of Educational Psychology, 89, 342-357.
Stanovich, K. E., & West, R. F. (1998a). Individual differences in
rational thought. Journal of Experimental Psychology: General,
127, 161-188.
Stanovich, K. E., & West, R. F. (1998b). Who uses base rates and
P(D/~H)? An analysis of individual differences. Memory &
Cognition, 28, 161-179.
Stanovich, K. E., West, R. F., & Harrison, M. (1995). Knowledge
growth and maintenance across the life span: The role of print
exposure. Developmental Psychology, 31, 811-826.
Stemberg, R. J. (1985). Beyond IQ: A triarchic theory of human
intelligence. Cambridge, England: Cambridge University Press.
Stemberg, R. J. (1997). The concept of intelligence and its role in
lifelong learning and success. American Psychologist, 52, 1030-
1037.
(Appendixes follow)
509
Stemberg, R. J., & Detterman, D. K. (1986). What is intelligence?
Norwood, NJ: Ablex.
Stigler, J. W., & Baranes, R. (1988). Culture and mathematics
learning. In E. Z. Rothkopf (Ed.), Review of Research in
Education (Vol. 15, pp. 253-306). Washington, DC: American
Educational Research Association.
Tabachnick, B. G., & Fidell, L. S. (1983). Using multivariate
statistics. New York: Harper & Row.
Troldahl, V., & Powell, F. (1965). A short-form dogmatism scale for
use in field studies. Social Forces, 44, 211-215.
Wade, C , & Tavris, C. (1993). Critical and creative thinking. New
York: HarperCollins.
Wasserman, E. A., Dorner, W. W, & Kao, S. F. (1990). Contribu-
tions of specific cell information to judgments of interevent
contingency. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 16, 509-521.
Webster, D. M., & Kruglanski, A. W. (1994). Individual differences
in need for cognitive closure. Journal of Personality and Social
Psychology, 67, 1049-1062.
Wechsler, D. (1981). Manual for Wechsler Adult Intelligence
Scale—Revised. New York: Psychological Corporation.
Wegener, D. T., & Petty, R. E. (1997). The flexible correction
model: The role of naive theories of bias in bias correction. In
M. P. Zanna (Ed.), Advances in experimental social psychology
(Vol. 29, pp. 141-208). San Diego, CA: Academic Press.
Wilson, T. D., & Brekke, N. (1994). Mental contamination and
mental correction: Unwanted influences on judgments and
evaluations. Psychological Bulletin, 116, 117-142.
Zechmeister, E. B., & Johnson, J. (1992). Critical thinking: A
functional approach. Pacific Grove: CA: Brooks/Cole.
Zimmerman, J., Broder, P. K., Shaughnessy, J. J., & Underwood,
B. J. (1977). A recognition test of vocabulary using signal-
detection measures, and some correlates of word and nonword
recognition. Intelligence, 1, 5-31.
 510 SA, WEST, AND STANOVICH

Appendix A
Items From Two Previously Unpublished Subscales
Items on the Flexible Thinking Subscale
1. Changing your mind is a sign of weakness. (Reflect)
2. A person should always consider new possibilities.
3. Intuition is the best guide in making decisions. (R)
4. If I think longer about a problem I will be more likely to solve it.
5. Basically, I know everything I need to know about the important things in life. (R)
6. Considering too many different opinions often leads to bad decisions. (R)
7. People should always take into consideration evidence that goes against their beliefs.
8. Difficulties can usually be overcome by thinking about the problem, rather than through waiting for good fortune.
9. There is nothing wrong with being undecided about many issues.
10. Coming to decisions quickly is a sign of wisdom. (R)
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Items on the Belief Identification Subscale

1. What beliefs you hold have more to do with your own personal character than the experiences that may have given rise to them.
2. It is a noble thing when someone holds the same beliefs as their parents.
3. One should disregard evidence that conflicts with your established beliefs.
4. Someone who attacks my beliefs is not insulting me personally. (R)
5. It is important to persevere in your beliefs even when evidence is brought to bear against them.
6. Certain beliefs are just too important to abandon no matter how good a case can be made against them.
7. Abandoning a previous belief is a sign of strong character. (R)
8. Beliefs should always be revised in response to new information or evidence. (R)
9. It makes me happy and proud when someone famous holds the same beliefs that I do.

Appendix B
Table Bl
Intercorrelations Among the Subscales on the Epistemic Regulation Questionnaire
Subscale 1 2 3 4 5 6 7
1. Flexible Thinking —
2. Openness-Ideas .250 —
3. Openness-Values .459 .171 —
4. Absolutism -.424 -.347 -.623 —
5. Dogmatism -.320 -.101 -.490 .513 —
6. Categorical Thinking -.254 -.060 -.456 .557 .550 —
7. Belief Identification -.527 -.245 -.523 .549 .378 .403
8. Need for Closure -.193 -.149 -.147 .356 .200 .167 .148
Note. Correlations larger than .176 are significant at the .05 level (two-tailed); correlations larger than .231 are significant at the .01 level
(two-tailed); correlations larger than .293 are significant at the .001 level (two-tailed).

Table B2
Component Loadings for all Subscales After Factor
Analysis and Varimax Rotation
Component
Subscale 1 2
Flexible Thinking -.588 .305
Openness-Ideas -.303 .312
Openness-Values -.728
Absolutism .833
Dogmatism .651
Categorical Thinking .667 .424
Belief Identification .690
Need for Closure .302
% variance accounted for 38.7% 6.1%
Note. Component loadings lower than .300 have been eliminated.

Received October 20,1998

Revision received December 15, 1998
Accepted February 4, 1999